Download - LASER AND MULTI-IMAGE REVERSE ENGINEERING ......methods for acquiring shape data, which use diverse mechanisms or phenomena to interact with the surface or volume of the object of

LASER AND MULTI-IMAGE REVERSE ENGINEERING SYSTEMS FOR ACCURATE

3D MODELLING OF COMPLEX CULTURAL ARTEFACTS

Ch. Ioannidis , G. Piniotis, S. Soile, F. Bourexis, A.-M. Boutsi, R. Chliverou, M. Tsakiri *

School of Rural and Surveying Engineering, National Technical University of Athens, Greece

KEY WORDS: 3D model, laser tracker, close range photogrammetry, reverse engineering

ABSTRACT:

The recent scientific and technical developments of reverse engineering methods and tools have broadened the possibilities of

applications in the field of cultural heritage conservation. In this paper, two different non-contact reverse engineering systems were

utilized for 3D data acquisition of a cultural heritage artefact. The object of interest is a 17th century wooden engraved ecclesiastical

sanctuary ciborium. The requirement of the 3D model is to aid the art conservators for the preservation of the wooden material and

the restoration of small damages and cracks in the engraved parts, thus requiring accuracy of the model in the order of sub-

millimetre. In this work, a Faro Vantage laser tracker was employed along with the FARO Edge Arm. In addition, image-based

modelling was also implemented with a large number of overlapping images acquired with a Canon EOS 6D camera and processed

using the well-known Structure from Motion (SfM) method with an auto-calibration procedure. The digital data acquisition and

processing procedures of the scanned geometry are described and compared to evaluate the performance of both systems in terms of

data acquisition time, processing time, reconstruction precision and final model quality. Whilst models produced with laser scanning

and image-based techniques is not a novel approach, the combination of laser tracking and photogrammetric data still presents

limited documentation in the field of cultural artefact documentation mainly due to the extremely high cost of the laser tracking

systems.

* Corresponding author

1. INTRODUCTION

Digital technologies as applied to relic conservation has become

a common research topic within the field of cultural heritage

preservation. Both the inheritance of traditional techniques and

the introduction of advanced technologies rely on the users and

their awareness and understanding of cultural heritage. It is

important to note that the implemented processes often combine

technology and art, and there are always new methods

appearing in the technological development. Digital technology

is currently one of all these strategies that inherits cultural

aspects, improves potency and raises quality. Whilst every

technology has advantages and limitations, it is necessary to

develop the benefits and avoiding the weaknesses and coming

up with possible and effective solutions.

One of the most accurate digital technologies is employed in

reverse engineering (RE). RE is the process of analysing an

object by deconstructing it to reveal its designs, architecture or

to extract knowledge from it, normally with the intention of

constructing a new object of similar or extended functionality.

The deconstruction is achieved in a digital sense by creating a

digital replica using very high accuracy 3D scanning devices.

RE is broadly utilized within various applications. For example,

in manufacturing (from analysis of an object to quality control

of industrial parts), in industrial design (creation of models for

virtual reality environments) and technical, industrial and

cultural heritage conservation, renovation, maintenance and

repair (e.g. Segreto et al., 2016; Bernard et al. 2007; Raja and

Fernandes 2008).

The development of RE methods has allowed for interesting

applications in the conservation of tangible cultural heritage

artefacts. One of the useful and modern applications is that of

creating a model for repair needs. RE techniques provide a

precious technological solution capable of acquiring the shape

and the advent of artefacts with great precision. Recent

advances in computer recourses and graphics capability have

expanded the visibility of possible damages in digital generated

cultural artefacts and RE is further used for inspection and

monitoring i.e. detection of artefact changes over the years (e.g.

Segreto et al., 2011).

1.1 Related work

The first step of the RE procedure is the digital data acquisition.

This is a critical stage of the RE procedure and clearly, the

choice of the RE data acquisition method affects the quality of

the acquired point cloud and, consequently, the resulting

reconstructed surface or CAD model. There are many different

methods for acquiring shape data, which use diverse

mechanisms or phenomena to interact with the surface or

volume of the object of interest. The main distinction is between

contact methods, that make use of devices such as mechanical

touch probes, and non-contact methods, commonly based on

light, sound or magnetic fields. Probably, the broadest and most

popular methods for geometrical data acquisition are the

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019 GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy

This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-2-W11-623-2019 | © Authors 2019. CC BY 4.0 License. 623

optical-based ones that offer reasonably fast acquisition rates

and precision. These are mainly the laser scanning systems and

the image-based techniques. Several experiences proposing RE

systems by non-contact and non-destructive techniques are

documented in the literature (e.g. Pruit et al., 2017).

In this work, the use of laser tracking technique is employed. Laser trackers are widely used for metrology and precision

surveys. Depending on the approach, range and instrument

itself, the measurement accuracy can vary from millimetre to

micron. The laser tracker is often accompanied by special

measuring arms and much of the success of this equipment is

tied to the laser tracker’s ability to measure with seven degrees

of freedom (DoF). In fact, there are six DoF that describe the

freedom of movement of a rigid body in 3D space and the 7th

DoF refers to the move through a larger work envelope (Martin,

2017). A laser tracker can gather information in three different

ways: by following a small- mirrored sphere, by tracking a

wireless, armless contact probe, or by tracking a handheld

scanner. The operator should choose the most appropriate data

acquisition method, or combination of tracking tools, for the job

at hand.

The most attractive alternative for data acquisition in RE is

photogrammetric image-based 3D modelling techniques which

have the major advantage of being low-cost, portable, flexible

and able to deliver highly detailed geometries and textures.

Objects with amorphous geometries, structured surfaces, many

edges, many corresponding image points and an

inhomogeneous colouring are best suited for this technique.

Objects that produce rather bad or no results have unstructured,

monochrome, translucent, reflective, and/or self-resembling

surfaces and most likely, cultural heritage artefacts fall in this

category thus, being problematic for image-based 3D modelling

(Nikolae et al., 2014; Schaich, 2013).

1.2 Aim of work

Nowadays, the combination of both photogrammetric and range

data (structured light, coded light or laser light) techniques are

used for 3D modelling when applied to cultural heritage to

enhance qualities of each one and achieve greater results (e.g.

Clini et al. 2016; Galo et al., 2013). There is however, limited

documentation on the use of RE laser tracking systems

combined with photogrammetric techniques on the

documentation of cultural artefacts. The aim of this paper is

therefore, to describe the rigorous workflow of the RE 3D

acquisition and modelling with non-invasive, high density and

of extremely high accuracy techniques for a case study of

cultural heritage artefact for conservation purposes. The object

of interest is a 17th century wooden engraved ecclesiastical

sanctuary ciborium of dimensions 1.40m x 1.40m x 3.00m (Fig.

1). The ciborium is a portable altar that is supported on four

columns/legs and is placed in the naves of Greek churches. It is

referred to as the ‘Holy Table’ and in Greek Orthodox churches

is always located in the centre of the altar. The specific

ciborium is situated in the church of Agios Stefanos in Meteora,

Greece and is currently used for the daily ecclesiastical services

of the monastery. The wooden material is exposed

uninterruptedly in localised temperatures due to the use of

candles and scents. In addition, the wood has been treated for

woodworms and currently it presents small cracks or missing

parts and it is within the priorities of the monastery to restore it.

Due to the complexity of this object, the most significant part of

the paper refers to overcoming the difficulties of acquisition and

the reconstruction of the texture. The remainder of the paper is

outlined as follows. In Section 2, the laser tracking pipeline is

described where the data acquisition methodology is presented

including a detailed description of the procedure (section 2.1).

Section 2.2 provides the data processing steps that have been

followed and the results. In Section 3, the digital image data

acquisition is described (section 3.1) and section 3.2 presents

the results of the processing steps. Section 4 provides a

comparison between the two types of point clouds and

polygonal meshes and section 5 summarises the main findings

of this work.

Fig.1 The ecclesiastical ciborium (17th cent.)

2. LASER TRACKING PIPELINE

The pipeline for scanning and virtual restoration processing to

form a texture mapped triangle mesh from sensed data is well

documented (eg. Bernadini et al., 2002). For RE manufacturing

applications, commercial products (e.g. Polyworks, Rapid-

Form) are available to convert point clouds to meshes. These

products also include some standard CAE/CAD analysis tools

such as measurements on surfaces and comparisons of as-built

to originally digitally designed parts.

However, the needs of cultural heritage applications are

different from traditional CAE/CAD in several different

respects. First, the objects of interest are typically not regularly

machined objects, but rather complex freeform shapes that were

produced manually and have been worn or broken over time

through natural processes. Second, while shape alone is

sometimes of interest, the material appearance of the object as

well as its shape are of interest. Third, several alteration

iterations are needed to evaluate the validity of a change. Whilst

in cultural heritage applications, judgments are often made on

the basis of appearance and matching images and textual

descriptions of appearance, in manufacturing RE it is important

to implement more readily quantified requirements such as

making parts fit together or minimizing weight. Fourth, the

population using digital models in cultural heritage is much


This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-2-W11-623-2019 | © Authors 2019. CC BY 4.0 License.

624

different from the one using CAE/CAD tools. While cultural

heritage specialists may have extensive technical training, their

primary interest is in the history of the objects being studied and

the people that used them and not in the technology itself.

Finally, the hardware and software resources available to

cultural heritage projects are generally less complex than those

of engineering design projects. In the following, a discussion on

the hardware used and the produced products is given.

2.1 Laser tracking system

The data acquisition was performed with the FARO ScanArm,

which is a Coordinate Measuring Machine (CMM) arm with an

attached hand-held laser scanner (Laser Line Probe, LLP) (cf.

Fig. 1). The ScanArm’s hard probe and the Laser Line Probe

can digitize interchangeably without having to remove either

component. The FARO LLP works via laser triangulation. The

LLP is, at its core, a digital camera that photographs laser light

and derives the depth based on the reflection return angle and

time. The LLP consists of a pulsing laser emitter that projects a

line of laser light, and a charge-coupled device (CCD) sensor,

which is a grid of pixels that convert photons of light into

electrical charges that are then interpreted as digital images

(www.faro.com). The CCD measures the reflection of the laser

photons emitted from the probe, and much like a camera, the

settings need to account for all of the things that can affect the

way that the photons will reflect back to the sensor. The arm has

an ergonomic pistol grip to enable the manual measurement of

3D points at any orientation within the arm’s spherical reach

(2.8 m), with a precision of ±0.040 mm and a repeatability of

±0.028 mm. The optical laser system mounted on the arm

allows to collect up to 560,000 points per second

(www.faro.com).

Prior to a measurement project, a compensation routine

procedure (sometimes referred to as calibration) needs to take

place. This routine is designed to closely control the position

and orientation of the LLP as the user covers multiple quadrants

of a compensation plate while moving the LLP. Specifically, the

operator takes a series of five strokes covering four positions of

the compensation plate while moving the LLP far from and near

to the plate. After compensation, it was necessary to set a

number of data acquisition settings in the software. Of the data

acquisition settings, there are three main variables that convey

resolution: the grid size, the maximum point distance (edge

length) and the maximum deviation. The grid size setting is

used to determine the average distance between points. As the

laser line passes across the surface of the material being

scanned, the sensor picks up more points than are necessary to

create an accurate model. For this project, it was set to 0.17mm.

The maximum edge length setting determines the distance

between adjacent laser stripes that will give continuous data.

The LLP collects data at a fixed rate, and if the probe is moved

too quickly, there will be gaps between the acquired data. The

maximum edge length determines how much overlap between

the scan lines is acceptable. A small maximum edge length

means that if the probe is moved too fast and the overlap is not

sufficient, the strips of data that are acquired will be discarded.

A larger maximum edge length means that less overlap is

necessary to preserve the data, and if there are gaps, they will be

filled automatically. This makes the scanning easier, but it also

means that the areas where no data points are collected will be

filled in with estimated data points, which diminishes the

accuracy of the model. The maximum edge length used for this

project was 3.5mm. The maximum deviation setting is a noise

reduction tool. Because the CCD sensor works by registering

the location of reflected laser light, sometimes photons with the

same energy signature as the laser being emitted from the LLP

enter the CCD sensor and register as data points. This ‘noise’ is

unavoidable, but it is undesired and therefore it must be

accounted for and eliminated if a scan is to meet standards of

accuracy for cultural heritage applications. One way to mitigate

this is by telling the software that if a data point comes in that

deviates from the rest of the data points being collected by a

certain distance, it will be discarded as noise. This maximum

deviation distance for this project was set to 0.05mm.

The second group of settings that needed to be defined deal

with the way the sensor actually acquires the data. There are

four basic settings that can be adjusted: noise threshold,

exposure, scan rate, and scan density. Of the four, only two are

commonly adjusted to account for the different colour and

reflectivity of an object: exposure and noise threshold. Both of

these can be automatically obtained by the software, much in

the way that automatic settings calculate exposure and the lens

aperture for a camera. The noise threshold setting configures the

sensitivity of the CCD to the returned photons; the electrical

charge of each pixel is measured on a scale of 0-255, and any

value below the noise threshold will be discarded

(www.faro.com). For the ciborium, the standard FARO-

recommended set point of 15 was used. The exposure setting for

the LLP controls the amount of time that the sensor collects

light per laser pulse, similar to how the exposure setting

controls the length of time for the shutter to be open on a digital

camera. The exposure is measured on a scale of 1-80, which

corresponds to an exposure time in milliseconds. For lighter-

coloured objects, a lower exposure setting is needed, and for

darker objects, a higher one. If a light object is scanned with too

high an exposure, the resulting scan will contain significant

noise. If a dark object is scanned with too low an exposure, the

electrical charge for each pixel on the CCD will be too low and

will be discarded. Because the wood is fairly dark, an exposure

setting of 70 was used, which corresponded to 4.375

milliseconds.

Finally, the last two settings are the scan rate and the scan

density. These settings control how much information is

obtained per laser pulse. The scan rate controls the number of

scan lines captured per second. The scan density controls the

number of points captured per line. These were left as the

default values in the software. The scanning arm has two

buttons; the green button, which starts the scanning and the red

that pauses it. During the scanning process, it is helpful to pause

and restart the scan often. If a scan is done too quickly or the

data does not meet the limits of the noise thresholds set prior to

scanning, scanned data will be discarded. The data is not

discarded on the fly though, it is discarded after the scanning is

either paused or stopped. Pausing allows the user to see which

parts of the scan are being kept and which parts are being

discarded. It is also a good idea to pause the scan prior to

repositioning the LLP; if the scanner is still trying to collect

data while the LLP is moved, it can scan something not

intended to be scanned, or collect noisy data.

Each scan was executed by manually following the surface

profile without restriction to a specific angle orientation; in this

way, data acquisition was extremely fast. From one position it

was not sufficient to acquire the entire object geometry

restricted by the arm’s length although the arm could be rotated

with a large freedom of movement. Therefore, scan overlapping

was necessary. The duration of data acquisition was in the order

of 5 hours for 8 different set-ups of the instrument. During the



625

scanning process, the laser probe has to be kept at a certain

distance from the object being scanned. If the probe is too close,

the laser cannot return to the sensor window, and likewise if it

is too far away. There is a range of distances that work

depending on the object being scanned. The LLP has a

rangefinder on it to show when the scanner is at the correct

distance from the object, and the computer screen will show

whether or not data is being collected in real time. Because of

the narrow range of distances at which data collection takes

place, it is especially important to pay close attention to the

screen when scanning the edges of the object at an angle

perpendicular to the edge. If the LLP is too far from the edge,

the tip of the edge can be in range but the faces not, or if the

LLP is held too close, the faces of the object can be in range but

the edge not. Since data acquisition is a product of the laser line

return to the sensor, it was necessary to scan from a variety of

angles to minimize shadow areas where the laser line return was

obstructed. Therefore, the most productive data acquisition

occurs when the LLP is held perpendicular to the face being

scanned, but for the engraved parts more oblique angles were

required to capture larger surface.

2.2 RE digital model

The digital model reconstruction of the acquired point clouds

was performed using the 3D metrology software platform

Polyworks V18 by InnovMetric (InnovMetric Software Inc.,

Quebec, Canada) (Reference Manual 2018). Polyworks is a

modularized package that offers a centralized workspace for

organizing component parts of 3D data processing and consists

of several modules, each dedicated to a specific phase of the RE

procedure, from scan alignment to meshing to inspection. The

Inspector module (IMSurvey) allows to scan the objects and

align the resulting datasets. The Modeler module covers several

essential steps for polygonal models editing, generation of

curves, non-uniform rational basis spline (NURBS) patches and

models that can be exported as CAD files readable by other

software tools. It also allows to compare the acquired data (e.g.

point cloud) with a reference (e.g. CAD model), measure the

dimensions of specific features and generate comparison and

verification reports.

Prior to any processing, registration (or alignment) of the

collected scans is essential to be performed. Within the

PolyWorks software, the different sets of point clouds are

entitled ‘Data objects’ and can be manually aligned to one or

several reference objects by pointing point pairs within each

data in the Align menu. An alignment is performed under the

Data alignments branch in ‘Tree View’. Following the

IMSurvey manual guide, the data alignments branch can be

described as a group of data alignments that refer to one or

multiple Data objects in data alignment groups. In general, data

alignment includes the position of point clouds after registration

process and also the parameters used and the type of alignment

carried out. Each of alignment performed is automatically added

to the list of Data alignment group, then it can be changed and

compared. The first stage of alignment is performed by selecting

at least three common points in the data objects using an initial

pairwise ICP (Iterative Closest Point) alignment. The second

stage is followed by a global registration step (such as best-fit

point-to-surface alignment, best fit cross sections, centre points

alignment etc) which improves the final alignment, spreading

the errors equally between all view pairs.

After alignment is completed, the point cloud is ready to

undertake further processing. In point cloud processing using

PolyWorks there are different operations available to improve

them: noise and overlap reduction, redundant points deletion.

The main PolyWorks tool for filtering data when the point

cloud is too dense is called Subsample. The tool offers random,

uniform and curvature-based sampling methods. For this

operation the point cloud needs to be visible and selected to

perform the procedure. The ‘Uniform’ method was chosen with

a tolerance value based on how much the point cloud needed to

be reduced.

The second phase involved the creation of the triangulation

mesh, the so-called ‘polygon mesh’. Within the software there

are three types of data triangulation: Triangulate Data Points,

Triangulate Terrain Data Points and Wrap Mesh around Point

Cloud. The first two are not suitable for this particular data set

and therefore the third algorithm was used. The ‘Wrap Mesh

around Point Cloud’ is probably the best method implemented

for creating a mesh because by this way the triangulation is

allowed to be best-fitted to a collection of points by subdividing

triangles and changing their vertices in 3D space. The module

offers additional options: rejection of outliers, noise reduction

and curvature-based subsampling. Fig. 2 shows the created

polygonal mesh of the ciborium using the ‘Wrap Mesh around

Point Cloud’ method.

Due to the high complexity of the object (free-form geometry) it

was not possible to perform further processing such as curves

and patches generation using NURBS construction.

Fig. 2 The 3D point cloud produced by the ScanArm

3. PHOTOGRAMMETRIC PIPELINE

The photogrammetric workflow comprised the following steps:

- Image acquisition with a digital camera;

- Camera calibration and image orientation, using a

Structure-from Motion (SfM) procedure able to

extract homologues points between the images and

the unknown camera parameters;

- Dense matching and 3D reconstruction for the

generation of dense point clouds;

- Polygonal model generation and texture mapping for

analyses and visualization.

3.1 Image acquisition

The photogrammetric acquisition was carried out in the altar

room of the church. A Canon 6D digital single-lens reflex

(DSLR) camera which employs a full frame CMOS sensor of

35.8 x 23.9mm, with resolution 20.2 effective megapixels and



626

6.5 μm pixel size as well as a Canon EF 24mm fixed focal

length lens providing an f1.4 maximum aperture. A set of 500

images was acquired with imaging acquisition distance of about

40cm.

Due to dim lights present in the room, external lamps were used

in order to avoid specular highlighting that can compromise the

photogrammetric process and produce a not faithful texture.

Thus, for the purposes of illumination, a Studio Flashlight

(1200 Watts) and its control panel was used combined with a

soft box of 30 x 40cm, for diffused lighting. This was essential

in order to achieve both smooth illumination and the use of the

smallest aperture value so as the largest depth of field to be

provided. The captures were made at 100ASA to ensure the

lowest digital noise level at the final produced photos.

A factor that cannot be underestimated during image data

acquisition is the availability of sufficient and safe space. The

limited space around the ciborium, its reasonable height and the

risk of hitting or damaging other assets when moving around

with bulky equipment was often stressful.

The acquired images obtained were processed through the

image-based 3D modelling software Agisoft PhotoScan. This

software is based on the structure-from-motion (SFM) and

dense multi-view 3D reconstruction (DMVR) algorithms, and

allows to build 3D models by unordered image collections that

depict a scene or an object from different viewpoints. First, the

software performs the alignment of the images, on the basis of

common points in the source photos and matches them to obtain

a single point cloud using a scale-invariant feature transform

approach. If necessary, the images were manually edited in

order to mask the background, leaving only the artefact visible

on the pictures. After image alignment, the generation of a

dense point cloud was automatically performed by the software

using a semi-global dense matching approach. The last step

was to generate a photorealistic texture on the 3D polygonal

model. The mesh model was opportunely edited to correct for

topological errors and the mesh model was textured with high

resolution external texture generated from the original images.

Clearly, the advantage of the SfM procedure is the

determination of internal camera geometry, position and

orientation automatically. The only cost is the required high

degree of overlap in image acquisition to cover the full

geometry of the object. However, when the object is complex,

as it the case in this work and the images are taken at short

distances, even the SfM operator cannot always guarantee the

precise orientation of images. An example is given in Fig. 3a

which depicts the automatically derived cloud from a façade of

the ciborium with problem areas mainly at the top of the object.

In order to assist the process, a number of homologous points

(tie-points) were manually selected on the images (about 30)

which facilitated in the calculation of the rotation and

translation parameters. Fig. 3b depicts the derived cloud using

manually measured tie-points. The alignment of the images was

performed with an accuracy of 0.1mm. Having the points clouds

derived from the two RE systems, the corresponding polygonal

surfaces (mesh) were created. Fig. 4a and 4b depict examples of

the two different surfaces. The former has an approximate

number of 3,000,000 triangles and the latter has about

22,000,000 triangles.

In order to compare and later combine the two cloud points for

further modelling (derived from SfM and laser scanning),

georeferencing was performed in a common coordinate system.

This was accomplished by ten points located on the four sides

of the ciborium and measured using the LLP at the quoted

precision of the manufacturer.

The final stage refered to the creation of the textured models by

providing external texture to the final mesh models and thus the

creation of the orthoimage. In the first case, the orthoimage was

derived using the SfM mesh (Fig. 5a) and in the second case the

orthoimage was derived using the laser scanning mesh (Fig. 5b).

The visual inspection of the two orthoimages indicates that the

SfM-derived one presents a complete model but the laser-

derived one has sharper details despite the occasional gaps.

Fig. 3a Point cloud from automatic SfM procedure without

manually introduced tie points

Fig. 3b Correct point cloud produced using SfM procedure with

manually measured tie points (about 4.000.000 points)

Fig. 4a Polygonal surface derived from the SfM procedure

Fig. 4b Polygonal surface derived from the scanning procedure



627

Fig. 5a Orthoimage derived using the DSM from the SfM

Fig. 5b Orthoimage derived using the DSM from scanning

4. RESULTS COMPARISON AND DICSUSSION

A number of comparisons were performed between the results

obtained with the two diverse RE systems: the first referred to

comparing the two point clouds, and the second referred to the

polygonal models generated from the point clouds. All the

comparisons were carried out considering the results obtained

with the laser arm system as ‘standard reference’. The

differences were computed via a shortest distance algorithm

between the two results and were represented as a coloured

map. For the comparisons, the CloudCompare open source

software was used.

For the point-to-point comparison, the two clouds (SfM and

laser scanning) were given as input in order to calculate their

difference. The coloured map of Fig. 6a depicts the maximum

distance, highlighted in colour blue, between the two types of

clouds with a mean of 1.28mm and standard deviation

±3.24mm. The comparison between the polygonal models (SfM

derived and laser scanning derived) is shown in the coloured

map of Fig. 6b with the maximum distance highlighted in

colour purple between the two types of surfaces with a mean of

1.33mm and standard deviation ±1.95mm. The mesh-to-mesh

volume comparison indicates differences of 2mm for more than

50% of matching cells (Fig. 6c).

From the comparisons, it can be noticed that the two models

obtained with the two diverse RE system are substantially

similar. The divergence between the two RE models is very low

but the real difference is given by the definition of the details.

For example, the same portion of the two polygonal models,

which is at around the centre part of the object, is shown to

visualize the high details definition. At the sides of the object

the differences are larger mainly from the SfM derived mesh.

The laser derived mesh presents a uniform precision.

In Table 1, a summary between the products derived from the

two RE systems (laser tracking and Digital Close-range

Photogrammetry –DCRP) is reported in terms of number of

acquired points, number of triangles, data acquisition time,

post-processing time, and RE system costs. The

photogrammetric system is of very low cost and faster than the

laser arm system, but the latter is more precise giving greater

details although with larger gaps due to the way the system is

operating (triangulation principle).

Fig. 6a Point-to-point comparison between the SfM and laser clouds

Fig. 6b Mesh-to-mesh comparison between the SfM and laser surfaces

Fig. 6c Mesh-to-mesh volume comparison between the SfM and laser

derived surfaces

Table 1. Comparison of the two RE models

Laser tracking DCRP

No of points 30,000,000 4,000,000

No of triangles 42,000,000 5,500,000

Acquisition time 5 hr 2 hr

Post-processing time 1 day 7 days

Cost Very high Low



628

5. CONCLUDING REMARKS

A case study of laser-based and multi image-based RE

procedures applied to a tangible cultural heritage artefact has

been illustrated by examining all the steps required to convert

the complex free-form geometry of a wooden engraved

ecclesiastical ciborium into its 3D digital representation.

On the basis of hardware features considerations, the laser arm-

based RE procedure, as expected, proved to be the most precise

but the produced cloud has larger number of gaps due to the

way the system operates (triangulation principle). Also, the

point cloud acquisition time was almost 2.5 times longer than

the phtogrammetric-based RE procedure. The digital

photogrammetric procedure is a mature technology and less

time consuming in the data acquisition stage.

The analysis of the obtained results takes into consideration not

only the accuracy of the final 3D model of the freeform object,

but also the required time to achieve this and the investment

relevance for system purchase. The point clouds derived from

the two systems are substantially different in terms of number of

points with the laser arm producing volumes of seven times

more than the photogrammetric process. The 3D digital models

of the object obtained with the two RE systems, were

comparatively evaluated via coloured maps defining the mean

distance between the two results. By this comparison, the 3D

polygonal surfaces derived from the two methods appear to be

similar with an overall mean distance of 1.33 mm. The derived

orthoimage using as DSM the laser point cloud has sharper

details but with larger number of gaps when compared to the

respective orthoimage produced using the SfM derived point

cloud.

Clearly, the investment cost for the laser arm-based system is

much higher than that for digital cameras. However, in view of

future developments, the availability of a precise 3D digital

model derived from both the two techniques as examined in this

work, could offer several opportunities within the cultural

heritage as well as the manufacturing fields.

ACKNOWLEDGEMENTS

The Monastery of Agios Stefanos in Meteora, Greece is

gratefully acknowledged for their permission and valuable help.

REFERENCES

Bernard, A., Laroche, F., Ammar-Khodja, S. Perry, N., 2007.

Impact of New 3D Numerical Devices and Environments on

Redesign and Valorisation of Mechanical Systems. CIRP

Annals – Manufacturing Technology 56 (1): 143–148.

Bernardini, F., and Rushmeier, H., 2002.The 3D Model

Acquisition Pipeline. Comput. Graph. Forum, 21_2_, pp. 149–

172.

Clini P., Frapiccini, N., Mengoni, M., Nespeca, R., Ruggeri, L.,

2014. SFM technique and focus stacking for digital

documentation of archaeological artifacts. Proc. ISPRS Volume

XLI-B5, 2016 XXIII ISPRS Congress, 12–19 July 2016,

Prague, Czech Republic.

Gallo, G., Muzzupappa, M., Bruno, F., 2014. 3D reconstruction

of small sized objects from a sequence of multi-focused images.

Journal of Cultural Heritage, Vol. 15(2), pp. 173-182.

Guidi, G., Rodríguez Navarro, O., Gonizzi-Barsanti, S., Micoli,

L.L., Russo, M., 2013. Quick textured mesh generation in

Cultural heritage digitization. Proc. Built Heritage 2013

Conference - Monitoring Conservation and Management,

pp.874-882.

Martin, J., 2017. The real-time promise of 7DoF laser tracking.

Quality Magazime, March, 2017(www.qualitymag.com).

Nicolae, C., Nocerino, E., Menna, F., Remondino, F., 2014.

Photogrammetry applied to problematic artefacts. Proc. ISPRS

Technical Commission V Symposium, Volume XL-5, 2014 23 –

25 June, Riva del Garda, Italy.

o Pruit, J.B., Clement, N.G., Tapanila, L., 2017. Laser and

structured light scanning to acquire 3D morphology. Virtual

Paleontology 22:57-69 https://doi.org/10.1017/scs.2017.8

Raja, V., and Fernandes K. J., eds. 2008. Reverse Engineering:

An Industrial Perspective. 242 p. London: Springer-Verlag.

Segreto, T., Caggiano, A. Teti, R., 2011. Diverse Non Contact

Reverse Engineering Systems for Cultural Heritage

Preservation. In 7th Int. Conf. on Digital Enterprise Technology

– DET 2011, 486–493, Sept. 28–30, Patras, ISBN 978-960-

88104-2-6.

Segreto, T., Bottillo, A., Teti, R., Galantucci, L.M., Lavecchia,

F., Galantucci, MB, 2016. Non-contact reverse engineering

modeling for additive manufacturing of down scaled cultural

artefact. In 10th CIRP Conf. on Intelligent Computation in

Manufacturing Engineering - CIRP ICME '16, 20-22 July,

Naples, Italy.

Schaich, M. 2013.Combined 3D scanning and photogrammetry

surveys with 3D database support for archaeology & cultural

heritage. A practice report on ArcTron’s information system

aSPECT3D. Proc. Photogrammetric Week 2013, D. Fritsch (ed.).

Wichmann, Berlin/Offenbach, pp. 233-246.



629

https://doi.org/10.1017/scs.2017.8

https://doi.org/10.1017/scs.2017.8

Download - LASER AND MULTI-IMAGE REVERSE ENGINEERING ......methods for acquiring shape data, which use diverse mechanisms or phenomena to interact with the surface or volume of the object of

Top Related